Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: Fwd: Questions
Nick Cox <email@example.com>
Re: st: Fwd: Questions
Sun, 4 Sep 2011 18:52:04 +0100
The analysis as conducted seems contradictory in its approach. If the
errors are heteroscedastic, their marginal distribution is not
expected to be a single Gaussian distribution. However, getting the
functional form right is much more important than getting the marginal
distribution of errors right.
A more modern approach, as in the blog entry just mentioned, is to use
a log link rather than a log transform and thus to use -poisson- or
some similar -glm-.
2011/9/4 Jorge Eduardo Pérez Pérez <firstname.lastname@example.org>:
> I think this email was intended to go to Statalist.
> Why not add a dummy variable for individuals in rural areas?
> Jorge Eduardo Pérez Pérez
> ---------- Forwarded message ----------
> From: Husaina Banu Kenayathulla <email@example.com>
> Date: Sun, Sep 4, 2011 at 9:08 AM
> Subject: Questions
> To: Perez Perez Jorge Eduardo <firstname.lastname@example.org>
> I am Husaina. I have some issues with my third analysis for my dissertation:
> I run the OLS regression separately for male and female labor force.
> My dependent variable is natural log of earnings (which includes
> earnings from self-employed and wage work). My independent variable
> is experience, exprience squared, years of scho0ling, ethnicity,
> urban/rural, and region.
> When I run the model and test for hetereskedasticity,my model violates
> the constant variance assumption for residuals. Thus, I use robust std
> erros. But, when I check for normality of residuals (using skewness
> and kurtosis test in STATA), it also violates the normality
> assumption. The q-q plot and p-plot is attached. If my normality
> assumption is violated, my estimates is unbiased but all my testing is
> invalid. I think the problem arise because I have large negative
> outliers but these cases are valid (individuals in rural areas whose
> income is very low). So, I can't drop them. They are also influential.
> I am not sure what to do. Do you have suggestions? Can I write the
> results when the normality assumption is violated?
* For searches and help try: